Shaping a Specialist: Skills and Duties

Skills, tools, duties

Three words come up constantly, and they are easy to mix up. A tool is a single action the employee can take, like sending an email or searching the web. A skill is packaged know-how: a repeatable way of doing a task, such as how you like a sales follow-up written, that the employee can apply whenever it fits. A duty is a standing responsibility it owns, like keeping your inbox triaged, something it returns to without being asked each time. The simplest way to hold them: a tool is what it can do, a skill is how it knows to do something well, and a duty is what it is on the hook for. You shape a specialist by choosing the right mix of all three, not by hoping one general setup covers everything.

• Tool (the action): one thing it can carry out, like "send an email".
• Skill (the know-how): a packaged workflow it applies, like "write a follow-up the way we do it".
• Duty (the responsibility): an ongoing job it owns, like "keep the pipeline updated".

Reusable instructions

The first time you explain how you want something done, you are briefing. The second time, you are repeating yourself. The fix is to save those instructions once so the employee carries them into every relevant task. That saved set of instructions is often called a playbook, and the always-on version that frames how the employee behaves is the system prompt. Write it the way you would brief a new hire who is good but new: what good looks like, what to avoid, the format you expect, the tone. Done once, it stops the re-explaining and makes the work come back consistent instead of slightly different every time.

Teaching it your standards

There are two ways to make an AI behave the way you want. The everyday one is giving it clear instructions and a few real examples of good work, and adjusting from there. That is configuration, and for almost everyone it is the whole job: it is fast, reversible, and you can change it any afternoon. The heavier one is fine-tuning: actually retraining the model on a large set of examples so the behaviour is baked in. It matters only when you have a narrow, high-volume task and a lot of clean examples, and even then it is slow and costly to redo. For shaping your employee around your work, reach for configuration first and treat fine-tuning as the rare exception.

• Configuration: instructions plus examples. Fast, reversible, what you will use almost always.
• Fine-tuning: retraining the model itself. Slow and costly, only for narrow high-volume cases.

Key takeaways

• A tool is an action, a skill is packaged know-how, a duty is a standing responsibility.
• Save how you want recurring work done as a playbook so you stop re-explaining it.
• Shape behaviour with instructions and examples (configuration), not retraining (fine-tuning).
• Reach for configuration first; fine-tuning is the rare exception, not the starting point.

Continue learning

• Working With AI Employees — View the complete course.
• Next: Putting It to Work: Automation and Teams